*/
/*
* This file is part of Lustre, http://www.lustre.org/
- * Lustre is a trademark of Sun Microsystems, Inc.
*
* lustre/mdt/mdt_lproc.c
*
#include <lustre_mds.h>
#include <lprocfs_status.h>
#include "mdt_internal.h"
+#include <obd_cksum.h>
/**
* The rename stats output would be YAML formats, like
* rename_stats:
- * - snapshot_time: 1234567890.123456
+ * - snapshot_time: 1234567890.123456789
+ * - start_time: 1234567880.987654321
+ * - elapsed_time: 9.135802468
* - same_dir:
* 4kB: { samples: 1230, pct: 33, cum_pct: 45 }
* 8kB: { samples: 1242, pct: 33, cum_pct: 78 }
**/
static void display_rename_stats(struct seq_file *seq, char *name,
- struct obd_histogram *hist)
-{
- unsigned long tot, t, cum = 0;
- int i;
-
- tot = lprocfs_oh_sum(hist);
- if (tot > 0)
- seq_printf(seq, "- %-15s\n", name);
- /* dir size start from 4K, start i from 10(2^10) here */
- for (i = 0; i < OBD_HIST_MAX; i++) {
- t = hist->oh_buckets[i];
- cum += t;
- if (cum == 0)
- continue;
-
- if (i < 10)
- seq_printf(seq, "%6s%d%s", " ", 1<< i, "bytes:");
- else if (i < 20)
- seq_printf(seq, "%6s%d%s", " ", 1<<(i-10), "KB:");
- else
- seq_printf(seq, "%6s%d%s", " ", 1<<(i-20), "MB:");
+ struct obd_histogram *rs_hist)
+{
+ unsigned long tot, t, cum = 0;
+ int i;
+
+ tot = lprocfs_oh_sum(rs_hist);
+ if (tot > 0)
+ seq_printf(seq, "- %s\n", name);
+
+ for (i = 0; i < OBD_HIST_MAX; i++) {
+ t = rs_hist->oh_buckets[i];
+ cum += t;
+ if (cum == 0)
+ continue;
+
+ if (i < 10)
+ seq_printf(seq, "%6s%d%s", " ", 1 << i, "bytes:");
+ else if (i < 20)
+ seq_printf(seq, "%6s%d%s", " ", 1 << (i - 10), "KB:");
+ else
+ seq_printf(seq, "%6s%d%s", " ", 1 << (i - 20), "MB:");
seq_printf(seq, " { sample: %3lu, pct: %3u, cum_pct: %3u }\n",
t, pct(t, tot), pct(cum, tot));
- if (cum == tot)
- break;
- }
+ if (cum == tot)
+ break;
+ }
}
static void rename_stats_show(struct seq_file *seq,
- struct rename_stats *rename_stats)
+ struct rename_stats *rename_stats)
{
- struct timespec64 now;
-
/* this sampling races with updates */
- ktime_get_real_ts64(&now);
- seq_printf(seq, "rename_stats:\n");
- seq_printf(seq, "- %-15s %llu.%9lu\n", "snapshot_time:",
- (s64)now.tv_sec, now.tv_nsec);
+ seq_puts(seq, "rename_stats:\n- ");
+ lprocfs_stats_header(seq, ktime_get(), rename_stats->rs_init, 15, ":",
+ false);
- display_rename_stats(seq, "same_dir",
- &rename_stats->hist[RENAME_SAMEDIR_SIZE]);
- display_rename_stats(seq, "crossdir_src",
- &rename_stats->hist[RENAME_CROSSDIR_SRC_SIZE]);
- display_rename_stats(seq, "crossdir_tgt",
- &rename_stats->hist[RENAME_CROSSDIR_TGT_SIZE]);
+ display_rename_stats(seq, "same_dir",
+ &rename_stats->rs_hist[RENAME_SAMEDIR_SIZE]);
+ display_rename_stats(seq, "crossdir_src",
+ &rename_stats->rs_hist[RENAME_CROSSDIR_SRC_SIZE]);
+ display_rename_stats(seq, "crossdir_tgt",
+ &rename_stats->rs_hist[RENAME_CROSSDIR_TGT_SIZE]);
}
static int mdt_rename_stats_seq_show(struct seq_file *seq, void *v)
{
- struct mdt_device *mdt = seq->private;
+ struct mdt_device *mdt = seq->private;
- rename_stats_show(seq, &mdt->mdt_rename_stats);
+ rename_stats_show(seq, &mdt->mdt_rename_stats);
- return 0;
+ return 0;
}
static ssize_t
mdt_rename_stats_seq_write(struct file *file, const char __user *buf,
size_t len, loff_t *off)
{
- struct seq_file *seq = file->private_data;
- struct mdt_device *mdt = seq->private;
- int i;
+ struct seq_file *seq = file->private_data;
+ struct mdt_device *mdt = seq->private;
+ int i;
- for (i = 0; i < RENAME_LAST; i++)
- lprocfs_oh_clear(&mdt->mdt_rename_stats.hist[i]);
+ for (i = 0; i < RENAME_LAST; i++)
+ lprocfs_oh_clear(&mdt->mdt_rename_stats.rs_hist[i]);
+ mdt->mdt_rename_stats.rs_init = ktime_get();
- return len;
+ return len;
}
LPROC_SEQ_FOPS(mdt_rename_stats);
int i;
for (i = 0; i < RENAME_LAST; i++)
- spin_lock_init(&mdt->mdt_rename_stats.hist[i].oh_lock);
+ spin_lock_init(&mdt->mdt_rename_stats.rs_hist[i].oh_lock);
return lprocfs_obd_seq_create(mdt2obd_dev(mdt), "rename_stats", 0644,
&mdt_rename_stats_fops, mdt);
struct mdt_device *mdt,
struct ptlrpc_request *req,
struct mdt_object *src,
- struct mdt_object *tgt)
+ struct mdt_object *tgt, long count)
{
- struct md_attr *ma = &info->mti_attr;
- struct rename_stats *rstats = &mdt->mdt_rename_stats;
- int rc;
+ struct md_attr *ma = &info->mti_attr;
+ struct rename_stats *rstats = &mdt->mdt_rename_stats;
+ int rc;
- ma->ma_need = MA_INODE;
- ma->ma_valid = 0;
- rc = mo_attr_get(info->mti_env, mdt_object_child(src), ma);
- if (rc) {
- CERROR("%s: "DFID" attr_get, rc = %d\n",
+ ma->ma_need = MA_INODE;
+ ma->ma_valid = 0;
+ rc = mo_attr_get(info->mti_env, mdt_object_child(src), ma);
+ if (rc) {
+ CERROR("%s: "DFID" attr_get, rc = %d\n",
mdt_obd_name(mdt), PFID(mdt_object_fid(src)), rc);
- return;
- }
-
- if (src == tgt) {
- mdt_counter_incr(req, LPROC_MDT_SAMEDIR_RENAME);
- lprocfs_oh_tally_log2(&rstats->hist[RENAME_SAMEDIR_SIZE],
- (unsigned int)ma->ma_attr.la_size);
- return;
- }
-
- mdt_counter_incr(req, LPROC_MDT_CROSSDIR_RENAME);
- lprocfs_oh_tally_log2(&rstats->hist[RENAME_CROSSDIR_SRC_SIZE],
- (unsigned int)ma->ma_attr.la_size);
-
- ma->ma_need = MA_INODE;
- ma->ma_valid = 0;
- rc = mo_attr_get(info->mti_env, mdt_object_child(tgt), ma);
- if (rc) {
- CERROR("%s: "DFID" attr_get, rc = %d\n",
+ return;
+ }
+
+ if (src == tgt) {
+ mdt_counter_incr(req, LPROC_MDT_SAMEDIR_RENAME, count);
+ lprocfs_oh_tally_log2(&rstats->rs_hist[RENAME_SAMEDIR_SIZE],
+ (unsigned int)ma->ma_attr.la_size);
+ return;
+ }
+
+ mdt_counter_incr(req, LPROC_MDT_CROSSDIR_RENAME, count);
+ lprocfs_oh_tally_log2(&rstats->rs_hist[RENAME_CROSSDIR_SRC_SIZE],
+ (unsigned int)ma->ma_attr.la_size);
+
+ ma->ma_need = MA_INODE;
+ ma->ma_valid = 0;
+ rc = mo_attr_get(info->mti_env, mdt_object_child(tgt), ma);
+ if (rc) {
+ CERROR("%s: "DFID" attr_get, rc = %d\n",
mdt_obd_name(mdt), PFID(mdt_object_fid(tgt)), rc);
- return;
- }
+ return;
+ }
- lprocfs_oh_tally_log2(&rstats->hist[RENAME_CROSSDIR_TGT_SIZE],
- (unsigned int)ma->ma_attr.la_size);
+ lprocfs_oh_tally_log2(&rstats->rs_hist[RENAME_CROSSDIR_TGT_SIZE],
+ (unsigned int)ma->ma_attr.la_size);
}
static ssize_t identity_expire_show(struct kobject *kobj,
*/
if (copy_from_user(kbuf, buf, min_t(unsigned long, BUFLEN - 1, count)))
GOTO(out, rc = -EFAULT);
- tmpbuf = cfs_firststr(kbuf, min_t(unsigned long, BUFLEN - 1, count));
+ tmpbuf = skip_spaces(kbuf);
+ tmpbuf = strsep(&tmpbuf, " \t\n\f\v\r");
if (strncmp(tmpbuf, "nid:", 4) != 0) {
count = lprocfs_evict_client_seq_write(file, buf, count, off);
return rc;
}
+ if (val == ALWAYS_DOM_LOCK_ON_OPEN)
+ val = TRYLOCK_DOM_ON_OPEN;
+
if (val < 0 || val >= NUM_DOM_LOCK_ON_OPEN_MODES)
return -EINVAL;
}
LUSTRE_RW_ATTR(dir_restripe_nsonly);
+static ssize_t enable_remote_subdir_mount_show(struct kobject *kobj,
+ struct attribute *attr,
+ char *buf)
+{
+ struct obd_device *obd = container_of(kobj, struct obd_device,
+ obd_kset.kobj);
+ struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
+
+ return scnprintf(buf, PAGE_SIZE, "%u\n",
+ mdt->mdt_enable_remote_subdir_mount);
+}
+
+static ssize_t enable_remote_subdir_mount_store(struct kobject *kobj,
+ struct attribute *attr,
+ const char *buffer,
+ size_t count)
+{
+ struct obd_device *obd = container_of(kobj, struct obd_device,
+ obd_kset.kobj);
+ struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
+ bool val;
+ int rc;
+
+ rc = kstrtobool(buffer, &val);
+ if (rc)
+ return rc;
+
+ mdt->mdt_enable_remote_subdir_mount = val;
+ return count;
+}
+LUSTRE_RW_ATTR(enable_remote_subdir_mount);
+
+/**
+ * Show if the OFD enforces T10PI checksum.
+ *
+ * \param[in] m seq_file handle
+ * \param[in] data unused for single entry
+ *
+ * \retval 0 on success
+ * \retval negative value on error
+ */
+static ssize_t checksum_t10pi_enforce_show(struct kobject *kobj,
+ struct attribute *attr,
+ char *buf)
+{
+ struct obd_device *obd = container_of(kobj, struct obd_device,
+ obd_kset.kobj);
+ struct lu_target *lut = obd->u.obt.obt_lut;
+
+ return scnprintf(buf, PAGE_SIZE, "%u\n", lut->lut_cksum_t10pi_enforce);
+}
+
+/**
+ * Force specific T10PI checksum modes to be enabled
+ *
+ * If T10PI *is* supported in hardware, allow only the supported T10PI type
+ * to be used. If T10PI is *not* supported by the OSD, setting the enforce
+ * parameter forces all T10PI types to be enabled (even if slower) for
+ * testing.
+ *
+ * The final determination of which algorithm to be used depends whether
+ * the client supports T10PI or not, and is handled at client connect time.
+ *
+ * \param[in] file proc file
+ * \param[in] buffer string which represents mode
+ * 1: set T10PI checksums enforced
+ * 0: unset T10PI checksums enforced
+ * \param[in] count \a buffer length
+ * \param[in] off unused for single entry
+ *
+ * \retval \a count on success
+ * \retval negative number on error
+ */
+static ssize_t checksum_t10pi_enforce_store(struct kobject *kobj,
+ struct attribute *attr,
+ const char *buffer, size_t count)
+{
+ struct obd_device *obd = container_of(kobj, struct obd_device,
+ obd_kset.kobj);
+ struct lu_target *lut = obd->u.obt.obt_lut;
+ bool enforce;
+ int rc;
+
+ rc = kstrtobool(buffer, &enforce);
+ if (rc)
+ return rc;
+
+ spin_lock(&lut->lut_flags_lock);
+ lut->lut_cksum_t10pi_enforce = enforce;
+ spin_unlock(&lut->lut_flags_lock);
+ return count;
+}
+LUSTRE_RW_ATTR(checksum_t10pi_enforce);
+
+/*
+ * mdt_checksum_type(server) proc handling
+ */
+DECLARE_CKSUM_NAME;
+
+static int mdt_checksum_type_seq_show(struct seq_file *m, void *data)
+{
+ struct obd_device *obd = m->private;
+ struct lu_target *lut;
+ enum cksum_types pref;
+ int i;
+
+ if (!obd)
+ return 0;
+
+ lut = obd->u.obt.obt_lut;
+
+ /* select fastest checksum type on the server */
+ pref = obd_cksum_type_select(obd->obd_name,
+ lut->lut_cksum_types_supported, 0);
+
+ for (i = 0; i < ARRAY_SIZE(cksum_name); i++) {
+ if ((BIT(i) & lut->lut_cksum_types_supported) == 0)
+ continue;
+
+ if (pref == BIT(i))
+ seq_printf(m, "[%s] ", cksum_name[i]);
+ else
+ seq_printf(m, "%s ", cksum_name[i]);
+ }
+ seq_puts(m, "\n");
+
+ return 0;
+}
+
+LPROC_SEQ_FOPS_RO(mdt_checksum_type);
+
LPROC_SEQ_FOPS_RO_TYPE(mdt, hash);
LPROC_SEQ_FOPS_WR_ONLY(mdt, mds_evict_client);
+LPROC_SEQ_FOPS_RW_TYPE(mdt, checksum_dump);
LUSTRE_RW_ATTR(job_cleanup_interval);
LPROC_SEQ_FOPS_RW_TYPE(mdt, nid_stats_clear);
LUSTRE_RW_ATTR(hsm_control);
&lustre_attr_dir_split_count.attr,
&lustre_attr_dir_split_delta.attr,
&lustre_attr_dir_restripe_nsonly.attr,
+ &lustre_attr_checksum_t10pi_enforce.attr,
+ &lustre_attr_enable_remote_subdir_mount.attr,
NULL,
};
.fops = &mdt_site_stats_fops },
{ .name = "evict_client",
.fops = &mdt_mds_evict_client_fops },
+ { .name = "checksum_dump",
+ .fops = &mdt_checksum_dump_fops },
{ .name = "hash_stats",
.fops = &mdt_hash_fops },
{ .name = "root_squash",
.fops = &mdt_root_squash_fops },
{ .name = "nosquash_nids",
.fops = &mdt_nosquash_nids_fops },
+ { .name = "checksum_type",
+ .fops = &mdt_checksum_type_fops },
{ NULL }
};
return 0;
}
-void mdt_counter_incr(struct ptlrpc_request *req, int opcode)
+void mdt_counter_incr(struct ptlrpc_request *req, int opcode, long amount)
{
struct obd_export *exp = req->rq_export;
if (exp->exp_obd && exp->exp_obd->obd_md_stats)
- lprocfs_counter_incr(exp->exp_obd->obd_md_stats,
- opcode + LPROC_MD_LAST_OPC);
+ lprocfs_counter_add(exp->exp_obd->obd_md_stats,
+ opcode + LPROC_MD_LAST_OPC, amount);
if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats != NULL)
- lprocfs_counter_incr(exp->exp_nid_stats->nid_stats, opcode);
+ lprocfs_counter_add(exp->exp_nid_stats->nid_stats, opcode,
+ amount);
if (exp->exp_obd && exp->exp_obd->u.obt.obt_jobstats.ojs_hash &&
(exp_connect_flags(exp) & OBD_CONNECT_JOBSTATS))
lprocfs_job_stats_log(exp->exp_obd,
lustre_msg_get_jobid(req->rq_reqmsg),
- opcode, 1);
+ opcode, amount);
}
static const char * const mdt_stats[] = {
[LPROC_MDT_IO_WRITE] = "write_bytes",
[LPROC_MDT_IO_PUNCH] = "punch",
[LPROC_MDT_MIGRATE] = "migrate",
+ [LPROC_MDT_FALLOCATE] = "fallocate",
};
-void mdt_stats_counter_init(struct lprocfs_stats *stats)
+void mdt_stats_counter_init(struct lprocfs_stats *stats, unsigned int offset)
{
- int idx;
-
- LASSERT(stats && stats->ls_num >= ARRAY_SIZE(mdt_stats));
+ int array_size = ARRAY_SIZE(mdt_stats);
+ int oidx; /* obd_md_stats index */
+ int midx; /* mdt_stats index */
- for (idx = 0; idx < ARRAY_SIZE(mdt_stats); idx++) {
- int flags = 0;
+ LASSERT(stats && stats->ls_num >= offset + array_size);
- if (idx == LPROC_MDT_IO_WRITE || idx == LPROC_MDT_IO_READ)
- flags = LPROCFS_CNTR_AVGMINMAX;
-
- lprocfs_counter_init(stats, idx, flags, mdt_stats[idx], "reqs");
+ for (midx = 0; midx < array_size; midx++) {
+ oidx = midx + offset;
+ if (midx == LPROC_MDT_IO_READ || midx == LPROC_MDT_IO_WRITE)
+ lprocfs_counter_init(stats, oidx,
+ LPROCFS_TYPE_BYTES_FULL,
+ mdt_stats[midx], "bytes");
+ else
+ lprocfs_counter_init(stats, oidx,
+ LPROCFS_TYPE_LATENCY,
+ mdt_stats[midx], "usecs");
}
}
{
struct obd_device *obd = mdt2obd_dev(mdt);
int rc;
- int i;
ENTRY;
LASSERT(name != NULL);
return rc;
/* add additional MDT md_stats after the default ones */
- for (i = 0; i < ARRAY_SIZE(mdt_stats); i++) {
- int idx = i + LPROC_MD_LAST_OPC;
- int flags = 0;
-
- if (idx == LPROC_MDT_IO_WRITE || idx == LPROC_MDT_IO_READ)
- flags = LPROCFS_CNTR_AVGMINMAX;
-
- lprocfs_counter_init(obd->obd_md_stats, idx, flags,
- mdt_stats[i], "reqs");
- }
-
+ mdt_stats_counter_init(obd->obd_md_stats, LPROC_MD_LAST_OPC);
rc = lprocfs_job_stats_init(obd, ARRAY_SIZE(mdt_stats),
mdt_stats_counter_init);